Statistical performance analysis of source localization in multisource, multiarray networks

With the advent of inexpensive, high-bandwidth networks, the use of multisensor and multiarray processing systems for distributed data collection is becoming possible. These new networks combine arrays of sensors with distributed sensor systems to create an "Array of Arrays." In this work, an array of arrays will be referred to as a Multiarray Network (MAN). MANs are being used to perform common signal processing tasks such as detection, source localization, and tracking.; In this thesis, a signal model that incorporates a field of multiple sources and a MAN is developed. The model is shown to reduce to previously published, less general models. Once a signal model is developed, the Cramer-Rao Bound for estimation of parameters in the model is derived. The Cramer-Rao Bound is lowest bound on the error variance of an unbiased estimate of parameters in a model. Mathematical analysis of the bound proves through Theorems 4.1-4.4 that increasing the number of sources, increasing the background noise and increasing the power of a source will increase the error variance of an estimated parameter of the model. Numerical analysis of the CRB shows that increasing the number of sources in a model is not equivalent to increasing the background noise of a model. When a new source is introduced into a model, the CRB must be fully re-derived to account for the effects of the new source.; Simulated data generated from the general models of multisource fields observed by multiarray networks is processed by algorithms to perform source localization. These algorithms include simple techniques like MUSIC DOA triangulations and G-MUSIC algorithms and more advanced signal processing including ICA DOA Triangulation and MI data association. Since the proposed source location estimation algorithms are complex and involve dynamically adaptive processing, Monte Carlo simulations are used to calculate statistical performance bounds of parameter estimation. The accuracy of each algorithm is found by comparison to the CRB. When the analysis is complete, it is shown that the ICA DOA Triangulation technique performs the worst and MI data association performs the best. The MUSIC based techniques all performed similarly with the G-MUSIC algorithm showing improvements over the other methods. Since G-MUSIC does not require a limited search for maxima, it is one of the easiest methods to implement.; The most significant, original contribution of this dissertation is the statistical performance analysis of Multisource/MAN source localization. The analysis provides both a theoretical lowest performance bound of MAN source localization and the statistical performance of 5 different source localization algorithms.